A fault diagnosis method and system for oil-immersed transformers, an electronic device, and a medium
Patent Information
- Application Number
- CN202311397365.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-25
- Publication Date
- 2026-10-27
- Estimated Expiration
- 2043-10-25
AI Technical Summary
[0005]本发明目的在于提供一种油浸式变压器故障诊断方法、系统、电子设备及介质，解决了现有方法中对油浸式变压器故障诊断可靠性和预测精度低的问题
[0031]本发明采用一种基于新的证据度量方法和置信Jensen–Shannon散度改进的D-S证据理论融合方法实现油浸式变压器故障诊断，针对单一模型学习不充分、预测精度低的局限，在传统D-S证据理论融合算法的基础上，实现了多个机器学习模型的信息融合并解决了多种模型预测结果不一致的问题，进而提高故障诊断的可靠性和精度。
Smart Images
Figure CN117407792B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of transformer fault diagnosis, specifically relating to a method, system, electronic equipment, and medium for diagnosing faults in oil-immersed transformers. Background Technology
[0002] In power systems, oil-immersed transformers are among the critical pieces of equipment, and their failures can lead to power system malfunctions and accidents. Therefore, timely and accurate fault diagnosis of oil-immersed transformers is of great importance in improving the reliability and stability of power systems.
[0003] Currently, some research results have been achieved in fault diagnosis of oil-immersed transformers. With the development of machine learning technology, researchers have used artificial intelligence models to diagnose early faults in oil-immersed transformers, improving the diagnostic speed. However, the performance of a single machine learning diagnostic model often suffers from low accuracy and reliability because it does not fully utilize the diverse learning capabilities of different machine learning methods.
[0004] DS evidence theory is an uncertainty reasoning theory and an extension of Bayesian theory. It can effectively handle uncertain information with unknown prior probabilities, compensating for the shortcomings of single machine learning diagnostic models. Traditional DS evidence theory can fuse evidence with low conflict; however, when high-conflict evidence exists, the conflict factor is 1, and the traditional DS evidence theory fusion rules fail. In practical problems, the outputs of machine learning algorithms may have high conflicts, thus traditional DS evidence theory may produce counterintuitive results, reducing the accuracy and reliability of fault diagnosis. Summary of the Invention
[0005] The purpose of this invention is to provide a method, system, electronic device and medium for fault diagnosis of oil-immersed transformers, which solves the problems of low reliability and prediction accuracy of fault diagnosis of oil-immersed transformers in existing methods.
[0006] The technical solution of the present invention is as follows:
[0007] A fault diagnosis method for an oil-immersed transformer, the method comprising:
[0008] Step 1: Obtain the fault dataset and normalize it. The dataset includes fault features and fault categories.
[0009] Step 2: Based on the dataset, obtain evidence using different machine learning algorithms, whereby the evidence is the Basic Probability Assignment Function (BPA).
[0010] Step 3: The improved DS evidence fusion algorithm is used to assign weights to the evidence to obtain a weighted average evidence.
[0011] Step 4: Calculate the weighted average evidence using the DS fusion rule to arrive at the final diagnostic result.
[0012] Furthermore, step one specifically includes:
[0013] Historical fault data of oil-immersed transformers were collected, and the data were processed by noise removal, outlier detection, and normalization.
[0014] Furthermore, step two specifically includes:
[0015] The dataset is divided into a training set and a test set. Different machine learning algorithms are pre-trained on the training set to obtain trained machine learning algorithms. The test set is input into the trained machine learning algorithms, and their output is converted into posterior probabilities to construct the basic probability assignment function BPA.
[0016] Furthermore, step three specifically includes:
[0017] Based on the evidence obtained in step two, and using the improved confidence Hellinger distance, the distance of each piece of evidence is first calculated, a distance matrix is constructed, and then the first weight factor of each piece of evidence is calculated from the distance matrix.
[0018] The confidence Jensen-Shannon divergence is used to measure the divergence between every two pieces of evidence to obtain the divergence matrix. Based on the divergence matrix, the average distance of each piece of evidence is determined, and the mutual support is determined by the average distance of each piece of evidence. Finally, the second weighting factor of each piece of evidence is obtained.
[0019] The credibility of each piece of evidence is determined by two weighting factors;
[0020] The weight of each piece of evidence is adjusted by the credibility rating to obtain a weighted average of the evidence.
[0021] Furthermore, step four specifically includes:
[0022] Using the combination rules of traditional DS evidence theory, the weighted average evidence is fused with itself N-1 times to obtain a matrix of one row and N columns, WAE. N-1 (E1), WAE N-1 (E2)...WAE N-1 (E i WAE N-1 (E n The numbers () represent fault types E1, E2...E respectively. i …E n The credibility of the evidence is determined, and the fault type with the highest credibility is taken as the output result; N is the number of pieces of evidence obtained in step two.
[0023] The present invention also provides a fault diagnosis system for oil-immersed transformers, the system comprising:
[0024] The data acquisition module is used to acquire a fault dataset and normalize the dataset, which includes fault characteristics and fault categories.
[0025] An evidence construction module is used to obtain evidence based on the dataset using different machine learning algorithms, wherein the evidence is a basic probability assignment function (BPA).
[0026] Weighting module: The improved DS evidence fusion algorithm is used to assign weights to the evidence to obtain a weighted average evidence.
[0027] Results generation module: Calculates the weighted average evidence using the DS fusion rule to arrive at the final diagnostic result.
[0028] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the oil-immersed transformer fault diagnosis method described above.
[0029] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for diagnosing faults in oil-immersed transformers.
[0030] Technical effects of the present invention:
[0031] This invention employs a novel evidence measurement method and a DS evidence theory fusion method based on confidence Jensen-Shannon divergence to achieve fault diagnosis of oil-immersed transformers. Addressing the limitations of insufficient learning and low prediction accuracy of a single model, this invention, based on the traditional DS evidence theory fusion algorithm, achieves information fusion of multiple machine learning models and solves the problem of inconsistent prediction results from various models, thereby improving the reliability and accuracy of fault diagnosis. Attached Figure Description
[0032] The accompanying drawings illustrate various embodiments generally by way of example rather than limitation, and are used, together with the specification and claims, to explain embodiments of the invention. Where appropriate, the same reference numerals are used in all drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the apparatus or method.
[0033] Figure 1 This is the overall algorithm flowchart of the present invention;
[0034] Figure 2 This is a comparison chart of the fault diagnosis accuracy of the present invention;
[0035] Figure 3 This is an example diagram of the fault diagnosis confusion matrix of the present invention. Detailed Implementation
[0036] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0037] This invention provides an example of fault diagnosis for oil-immersed transformers, the process of which is as follows: Figure 1 As shown, the experimental data comes from relevant literature. Data from 324 oil-immersed transformers were used, with five fault types (E1, E2, E3, E4, E5). Each fault type has five characteristics: hydrogen (H2), methane (CH4), ethane (C2H6), ethane (C2H4), and acetylene (C2H2). 226 randomly selected data points containing each fault type and characteristic were used to train five different machine learning models: K-Nearest Neighbors (KNN), Gradient Boosting (GB), Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM). The remaining 98 data points were used as test samples. The test set was selected to illustrate the implementation steps of the proposed fault diagnosis method.
[0038] Step 1: Collect fault sample data of oil-immersed transformers from relevant literature, and normalize the data. The normalization formula is as follows:
[0039]
[0040] The fault sample data has five features and five fault types. The identification framework Ω = {E1, E2, ..., E5} is a set of fault types. p ∈X={x1,…,x p ,…,x m} is a sample of the fault types to be diagnosed in the test set.
[0041] Step 2: Divide the dataset into a training set and a test set, accounting for 70% and 30% of the total dataset, respectively. The training set is used to train the model, and the test set is used to evaluate the model. The confusion matrix is used as the model evaluation metric. Figure 2 As shown.
[0042] A classifier is used to diagnose the fault types in the test set, and its output is converted into a posterior probability, thereby constructing the basic probability assignment function BPA. BPA is evidence, and there are five pieces of evidence in total.
[0043] Step 3: Using a new evidence measurement method and a DS evidence fusion algorithm improved by confidence Jensen-Shannon divergence, information from different classification models is reasonably weighted;
[0044] Based on the N=5 pieces of evidence constructed in step two, the distance for each piece of evidence is first calculated using the improved confidence Hellinger distance, as shown in the following formula:
[0045]
[0046] Where c = 1 / n, n = 5 represents the total number of fault types, m i (A j The distance H represents the probability that the sample belongs to the j-th fault type after passing the i-th algorithm. Finally, a 1×N distance matrix H = [H1, H2, ..., Hj] is constructed for each test sample. N ];
[0047] Secondly, from the distance matrix H = [H1, H2, ..., H N ] Calculate the first weighting factor for each piece of evidence:
[0048]
[0049] The Jensen-Shannon divergence is used to measure the divergence between every two pieces of evidence. The formula is as follows:
[0050]
[0051] in
[0052] Thus, the N×N scatter matrix is obtained as follows:
[0053]
[0054] In the above formula, BJS ij =BJS ji BJS ii =0, therefore matrix D is a symmetric matrix with 0 on the main diagonal;
[0055] Based on the divergence matrix D, the average distance for each piece of evidence is determined as follows:
[0056]
[0057] The degree of mutual support is determined by the average distance between each piece of evidence:
[0058]
[0059] Calculate the second weighting factor for each piece of evidence based on the divergence matrix D:
[0060]
[0061] The credibility of each piece of evidence was determined using two weighting factors. i for:
[0062] Crd i = w1(i) × w2(i)
[0063] Based on credibility Crd i After adjusting the weights of each piece of evidence, the weighted average evidence (WAE) is obtained, calculated using the following formula:
[0064]
[0065] Step four specifically includes:
[0066] There are five fault types: E1, E2, E3, E4, and E5. Using the combination rules of traditional DS evidence theory, the weighted average evidence WAE is fused with itself, resulting in the first fusion result, WAE1:
[0067]
[0068]
[0069] Where B∈2 Ω C∈2 Ω K∈[0,1] is the conflict coefficient between the evidence. The larger K is, the greater the difference between the evidence. If K=1, it means that the two pieces of evidence are completely conflicting. This rule only applies when K<1. Then, the first fusion result WAE1 and the weighted average evidence WAE are fused to obtain the second fusion result WAE2:
[0070]
[0071] Where B∈2 Ω C∈2 Ω WAE is the weighted average evidence. WAE1 is the first fusion result, which is the basic probability assignment value after fusing WAE with WAE. Then, the second fusion result WAE2 and the weighted average evidence WAE are fused to obtain the third fusion result WAE3.
[0072]
[0073] Where B∈2 Ω C∈2 ΩWAE is the weighted average evidence, and WAE2 is the second fusion result, which is the basic probability assignment value after fusing WAE1 and WAE. Then, the third fusion result WAE3 and the weighted average evidence WAE are fused to obtain the fourth fusion result WAE4.
[0074]
[0075] Where B∈2 Ω C∈2 Ω WAE is the weighted average evidence, and WAE3 is the third fusion result, which is the basic probability assignment value after the fusion of WAE2 and WAE.
[0076] After merging N-1=4, a judgment is made. WAE4 is a matrix with one row and five columns. WAE4(E1), WAE4(E2), WAE4(E3), WAE4(E4), and WAE4(E5) represent fault types E1, E2, and E5, respectively. Experimental verification:
[0077] To verify the effectiveness of the algorithm proposed in this invention, the fusion rules based on Jousselme evidence distance (Y. Deng method), the fusion rules based on confidence Jensen-Shannon divergence and information entropy (FYXiao method), and the fusion rules based on confidence Hellinger distance and information entropy (CSZhu et al. method) are compared with the algorithm proposed in this invention.
[0078] Firstly, to verify the improvement effect of the algorithm of this invention on the fault diagnosis accuracy of oil-immersed transformers, the diagnostic accuracy was compared with that of five basic classification algorithms and three different evidence theory fusion methods. Figure 3 It is evident that the method proposed in this invention achieves the highest diagnostic accuracy. This is because a single classification model cannot fully learn the relationship between features and fault types, and discrepancies exist. DS evidence theory mitigates the conflict between different classification models and improves fault diagnosis accuracy, but it has certain limitations. In contrast, the DS evidence theory fusion method based on a novel evidence measurement method and confidence Jensen-Shannon divergence fully utilizes all information about the evidence and updates the weights of the evidence.
[0079] To further verify the optimization effect of the improved algorithm on fault diagnosis, the recall rate and F1 score of the eight algorithms were compared, as shown in Table 1 below:
[0080] Table 1: Comparison of recall and F1 score for different algorithms
[0081] K-neighbor 82.65 0.8427 Gradient boosting 89.80 0.9013 Random Forest 88.78 0.9019 Decision Tree 84.69 0.8372 Support Vector Machine 68.37 0.7413 Y.Deng 91.84 0.9195 FYXiao 91.84 0.9312 CSZhu et al. 91.84 0.9312 The proposed method 93.88 0.9413
[0082] In the table above, compared with other algorithms, the evidence fusion algorithm of this invention has the highest recall and F1 score, which indicates that the improved algorithm has better overall performance and reflects the effectiveness of the method.
[0083] The present invention also provides a fault diagnosis system for oil-immersed transformers, the system comprising:
[0084] The data acquisition module is used to acquire a fault dataset and normalize the dataset, which includes fault characteristics and fault categories.
[0085] An evidence construction module is used to obtain evidence based on the dataset using different machine learning algorithms, wherein the evidence is a basic probability assignment function (BPA).
[0086] Weighting module: The improved DS evidence fusion algorithm is used to assign weights to the evidence to obtain a weighted average evidence.
[0087] Results generation module: Calculates the weighted average evidence using the DS fusion rule to arrive at the final diagnostic result.
[0088] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the oil-immersed transformer fault diagnosis method described above.
[0089] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for diagnosing faults in oil-immersed transformers.
[0090] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A fault diagnosis method for an oil-immersed transformer, characterized in that, The method includes: Step 1: Obtain the fault dataset and normalize it. The dataset includes fault features and fault categories. Step 2: Based on the dataset, obtain evidence using different machine learning algorithms, whereby the evidence is the Basic Probability Assignment Function (BPA). Step 3: The improved DS evidence fusion algorithm is used to assign weights to the evidence, resulting in a weighted average evidence; specifically, this includes: Based on the evidence obtained in step two, and using the improved confidence Hellinger distance, the distance of each piece of evidence is first calculated, a distance matrix is constructed, and then the first weight factor of each piece of evidence is calculated from the distance matrix. The distance calculation formula for each piece of evidence is as follows: in c = 1 / n , n= 5 indicates the total number of fault types. Indicates passing through the first i The algorithm, the sample belongs to the... j The probability of each fault type is ultimately used to construct a [probability] for each test sample. Distance matrix The confidence Jensen-Shannon divergence is used to measure the divergence between every two pieces of evidence to obtain the divergence matrix. Based on the divergence matrix, the average distance of each piece of evidence is determined, and the mutual support is determined by the average distance of each piece of evidence. Finally, the second weighting factor of each piece of evidence is obtained. The credibility of each piece of evidence is determined by two weighting factors; The weight of each piece of evidence is adjusted by its credibility to obtain a weighted average of the evidence. Step 4: Calculate the weighted average evidence using the DS fusion rule to arrive at the final diagnostic result; specifically including: Using the combination rules of traditional DS evidence theory, the weighted average evidence is fused with itself N-1 times to obtain a single line. N A matrix of columns, wherein the parameters of the matrix include: , … … These represent the types of faults. E 1. E 2… E i … E n The credibility of the evidence is determined, and the fault type with the highest credibility is taken as the output result; N is the number of pieces of evidence obtained in step two.
2. The fault diagnosis method for oil-immersed transformers according to claim 1, characterized in that, Step one specifically includes: Historical fault data of oil-immersed transformers were collected, and the data were processed by noise removal, outlier detection, and normalization.
3. The fault diagnosis method for oil-immersed transformers according to claim 1, characterized in that, Step two specifically includes: The dataset is divided into a training set and a test set. Different machine learning algorithms are pre-trained on the training set to obtain trained machine learning algorithms. The test set is input into the trained machine learning algorithms, and their output is converted into posterior probabilities to construct the basic probability assignment function BPA.
4. A fault diagnosis system for an oil-immersed transformer, using the method as described in any one of claims 1-3, characterized in that, The system includes: The data acquisition module is used to acquire a fault dataset and normalize the dataset, which includes fault characteristics and fault categories. An evidence construction module is used to obtain evidence based on the dataset using different machine learning algorithms, wherein the evidence is a basic probability assignment function (BPA). Weighting module: The improved DS evidence fusion algorithm is used to assign weights to the evidence to obtain a weighted average evidence. Results generation module: Calculates the weighted average evidence using the DS fusion rule to arrive at the final diagnostic result.
5. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the oil-immersed transformer fault diagnosis method according to any one of claims 1-3.
6. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the oil-immersed transformer fault diagnosis method as described in any one of claims 1-3.